Thanks for the feedback. I have moved a version of sanei_init_debug into hpaio source tree. This should remove the libsane.so dependency in the next hplip release.
-dave > -----Original Message----- > From: [EMAIL PROTECTED] > [mailto:[EMAIL PROTECTED] On Behalf > Of Jeremy White > Sent: Monday, March 19, 2007 6:21 AM > To: hplip-devel@lists.sourceforge.net > Subject: [HPLIP-Devel] Subtle bug with libsane-hpaio.so and > unresolvedsymbol sanei_init_debug > > I recently uncovered what appears to be a subtle bug with > libsane-hpaio.so. I did some digging, but felt that I wasn't > familiar enough with either sane or the hplip project to > propose a patch. > > I'm hoping that someone here will step up and resolve the issue. > > Specifically, libsane-hpaio.so is built with a weak symbolic > reference to sanei_init_debug; all 'normal' sane backends > include their own private copy of that function and so have > no need for libsane.so. This breaks applications in rare situations. > > Now, in normal operation, libsane is dlopen'd by an > application with RTLD_GLOBAL, so that all of it's symbols are > available for further resolution. > > When libsane initializes, it dlopen's libsane-hpaio with > local|lazy, and then dlsyms to the various functions. Then, > when those functions are invoked, ld fixes them up, uses the > global libsane to resolve sanei_init_debug, and life is good. > > However, if you have an application which dlopen's libsane.so > *without* RTLD_GLOBAL, life is much worse. Because when ld > goes to fix up the dlsym'd functions, it realizes - ZOMG - > that sanei_debug_init isn't available, and crashes the whole > application. > > This results in Wine applications crashing any time they > attempt to initialize libsane, which is quite awful, as you > can imagine. > > Now, I've patched Wine to explicitly load libsane with RTLD_GLOBAL: > > http://source.winehq.org/git/wine.git/?a=commit;h=16f33d8b436a > abf3b9e3ba810b68b596d9ddd687 > > But I felt honor bound to report this problem 'upstream', in > the hopes that someone here could fix it properly. At the > least, I've now done my duty :-/. > > I've attached a test program I used in probing this, in case > it's handy, along with output runs (on a Fedora Core 6 box, > where I first noticed this) that illustrate the problem. > > Cheers, > > Jeremy > > ------------------------------------------------------------------------- Take Surveys. Earn Cash. Influence the Future of IT Join SourceForge.net's Techsay panel and you'll get the chance to share your opinions on IT & business topics through brief surveys-and earn cash http://www.techsay.com/default.php?page=join.php&p=sourceforge&CID=DEVDEV _______________________________________________ HPLIP-Devel mailing list HPLIP-Devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/hplip-devel
